catch them coming and going: probabilistic pathway ... · in start’s analysis by running a...

National Consortium for the Study of Terrorism and Responses to Terrorism A Department of Homeland Security Science and Technology Center of Excellence

Led by the University of Maryland

8400 Baltimore Ave., Suite 250 • College Park, MD 20742 • 301.405.6600

www.start.umd.edu

Catch Them Coming and Going: Probabilistic Pathway Projection to Counter Foreign Fighters Final Research Report

National Geospatial-Intelligence Agency

March 2016

http://www.start.umd.edu/

National Consortium for the Study of Terrorism and Responses to Terrorism

A Department of Homeland Security Science and Technology Center of Excellence

1

About This Report

The authors of this report are Marcus A. Boyd, GIS Senior Researcher at the National Consortium for the

Study of Terrorism and Responses to Terrorism (START), Joshua M. House, GIS Analyst at START, Brady

L. Woods, GIS Analyst at START, Douglas I. Zietz, GIS Researcher Assistant at START and Rhys A. Young,

GIS Researcher Assistant at START. Questions about this report should be directed to Marcus A. Boyd at

[email protected].

This report is part of the National Consortium for the Study of Terrorism and Responses to Terrorism

(START) project, “Catch Them Coming and Going: Probabilistic Pathway Projection to Counter Foreign

Fighters,” led by Marcus A. Boyd.

This research was supported by the National Geospatial Intelligence Agency through Award Number

HM047615C0017, Project Number NA88G15169AS04. The views and conclusions contained in this

document are those of the authors and should not be interpreted as necessarily representing the official

policies, either expressed or implied, of the National Geospatial Intelligence Agency, U.S. Department of

Homeland Security or START.

About START

The National Consortium for the Study of Terrorism and Responses to Terrorism (START) is supported in

part by the Science and Technology Directorate of the U.S. Department of Homeland Security through a

Center of Excellence program led by the University of Maryland. START uses state‐of‐the‐art theories,

methods and data from the social and behavioral sciences to improve understanding of the origins,

dynamics and social and psychological impacts of terrorism. For more information, contact START at

[email protected] or visit www.start.umd.edu.

Citations

To cite this report, please use this format:

Marcus A. Boyd, Joshua M. House, Brady L. Woods, Douglas I. Zietz, and Rhys A. Young. “Catch Them

Coming and Going: Probabilistic Pathway Projection to Counter Foreign Fighters,” Final Research Report

to NGA. College Park, MD: START, 2016.

mailto:[email protected]

http://www.start.umd.edu/



2

Contents

Project Background and Introduction ....................................................................................................................................... 3

Analytical Methodology .................................................................................................................................................................. 4

Data ......................................................................................................................................................................................................... 4

Building Transportation Networks ............................................................................................................................................ 5

Intermodal Network Transitions .............................................................................................................................................. 10

Areas of Operation .......................................................................................................................................................................... 11

Analysis and Results ....................................................................................................................................................................... 19

Scenario 1: ..................................................................................................................................................................................... 21

Scenario 2: ..................................................................................................................................................................................... 22

Scenario 3: ..................................................................................................................................................................................... 23

Scenario 4: ..................................................................................................................................................................................... 23

Scenario 5: ..................................................................................................................................................................................... 24

Scenario 6: ..................................................................................................................................................................................... 24

Scenario 7: ..................................................................................................................................................................................... 25

Scenario 8: ..................................................................................................................................................................................... 26

Scenario 9: ..................................................................................................................................................................................... 26

Scenario 10: .................................................................................................................................................................................. 27

Scenario 11: .................................................................................................................................................................................. 28

Scenario 12: .................................................................................................................................................................................. 28

Scenario 13: .................................................................................................................................................................................. 29

Scenario 14: .................................................................................................................................................................................. 30

Research Opportunity ................................................................................................................................................................... 31

Conclusions ........................................................................................................................................................................................ 31



3

Project Background and Introduction

The Analysis and Production (A) Division of the National Geospatial Intelligence Agency (NGA) identified

a requirement for an assessment of the possible transportation corridors used by foreign fighters when

traveling to and from Islamic State of Iraq and Levant (ISIL), specifically focusing on transportation

networks in Europe, North Africa, and the Levant. As part of this effort, START completed qualitative and

quantitative research to assess the risk and the threats posed by foreign fighters in regards to

international violence and influence. START also conducted geospatial transportation analyses to identify

possible and most probable routes that could potentially support both licit and illicit movement of

foreign fighters as well as modes of transport that could be used in such activities.

START used methods and analysis from the Transnational Illicit Trafficking (TransIT) tool1 to create a

subset of the larger model for the specific purpose of developing a scientifically grounded methodology

for identifying the most likely routes of foreign fighters into and out of conflict zones. The complexity of

the multi-modal network is specifically designed to model all likely modalities used by a foreign fighter.

The tool includes detailed modeling of illicit pathways known to be used by foreign fighters as well as a

traditional licit means of routing. START’s analysis used Dijkstra's algorithm in ArcGIS’s Network Analyst

to determine optimal routes for traveling throughout the study area.

The unexpected flood of refugees and economic migrants into Europe and the subsequent introduction of

security measures created a significant alteration to our initial analysis plan. This has been accounted for

in START’s analysis by running a baseline analysis prior to the introduction of border measures as a

control test and running analysis with these measures in place to show potential deviation in route

planning. The analysis incorporates the added complexity of increased security measures in Europe and

integrates the rampant human trafficking between Turkey and Greek islands in the Aegean.

The following report provides a methodology and analysis of transit between Europe and ISIL-held

territory. START examines modeled routes originating in Europe and ending in Syria and Iraq and vice

versa. Assessing the likelihood of a chosen route requires a consideration of the interaction between

time, speed, and choice of modality. The selection of specific routes by the potential foreign fighters is a

product of both the shortest path distance between two nodes and a number of potential costs and

benefits associated with a traversed route.

This report begins with an analytic methodology describing the methods used to compile, construct, and

run the model. The Data section describes the data sources, various networks, and weights in detail. The

1 The TransIT Tool was originally designed for the TCOTRN Project, a similar study to the current study conducted in the Central and South American regions, in 2012. Please refer to Behlendorf, Brandon. Geospatial Modeling of Potential Routes for RN Trafficking: Central America and the Caribbean Basin. Interim Report to the Domestic Nuclear Detection Office. College Park, MD: START, 2012, for more detail on the origin of the TransIT Tool and TCOTRN specific route analysis.



4

Analysis and Results section provides a comprehensive investigation of the model’s results under

differing parameters. The section on Research Opportunities provides an overview of potential growth of

the project.

Analytical Methodology

In order to analyze probable routes for foreign fighters entering ISIL-controlled territory, we use an

algorithm originally developed by Dijkstra2 with custom parameters. This algorithm is applied to a multi-

modal transportation network to generate probable routes to and from ISIL-controlled territory. The

integrated multi-modal transportation network across the study region is a bidirectional graph that

consists of a set of vertices and edges. The total network consists of over 1.4 million edges, 100 points in

Europe and North Africa, and 100 points located near or within ISIL-controlled areas. The points in

Europe and North Africa consist of major cities such as Brussels, Berlin, Copenhagen, Paris, London,

Budapest, Milan, Casablanca, Stockholm, and Tunis. Of the 100 points located near or in ISIL-controlled

area, 52 of them were randomly generated, while the other 48 were cities of interest such as Ramadi,

Tikrit, Al Fallujah, Kirkuk, Mosul, Ar Raqqah, Tamdur, Abu Kamal, Aleppo, and Damascus.

If foreign fighters want to minimize risk when traveling to or from ISIL-controlled areas, the baseline

assumption of this model is that they will choose the least cost path from an origin point in Europe to an

ISIL-controlled destination (or vice versa). The unweighted cost variable, speed, is applied to the edges in

order to weight the graph. The weighted cost variable is a function of time (derived from distance and

speed) and risk. A potential foreign fighter will prefer to take the shortest pathway with the lowest

amount of risk. To this end, the model is a simple function of distance (measured in time to traverse a

segment) weighted by the costs and incentives discussed below.

Data

START began by creating a conventional licit transportation network for the Europe, North Africa, and

Levant regions. The licit network includes primary modes of transportation used by individuals when

travelling domestically or internationally across the study region. These modes of transportation include

roadways, rail lines, air lines, and maritime ferry routes. The network then adds on an illicit mode of

transportation that individuals could exploit to avoid detection when travelling to or from the Islamic

State operational areas. This primarily focuses on maritime routes used by both migrants and foreign

fighters exiting the Turkish mainland via the Aegean Sea in hopes of seeking asylum in the European

Union.

2 Dijkstra, E.W. "A Note on Two Problems in Connexion with Graphs." Numerische Mathematik http://www-m3.ma.tum.de/foswiki/pub/MN0506/WebHome/dijkstra.pdf (accessed November 19, 2014)



5

Building Transportation Networks Aerial Network

START collected the flight information from the Federal Aviation Administration's (FAA) Flight

Database.3 The data includes all flights for a single month as START recognizes that commercial airline

carriers fly on regular schedules that repeat daily, weekly, and monthly. This data is not comprehensive

but represents a substantial number of routes (fig. 1) providing an acceptable representative model for

both commercial and civil (corporate jets) aviation. In addition to the airline flight data, the origin and

destination airport locations were also derived from the FAA dataset.

Fig. 1 Aerial Network

Land Network

Roadways

The road network (fig. 2) is comprised of highway, primary, secondary, tertiary, local/urban, trail, and

unspecified roads derived from an open source global roads file provided by the National Aeronautics

and Space Administration (NASA) Socioeconomic Data and Applications Center (SEDAC).4 Although this

dataset lacks directional attributes and Z elevation usually found in the network models of city-level

transportation planning, considering the regional-level focus of the model, the absence of these elements

3 National Transportation Atlas Database 2011, Bureau of Transportation Statistics, Office of the Assistant Secretary for Research and Technology. 4 Center for International Earth Science Information Network - CIESIN - Columbia University, United Nations Food and Agriculture Programme - FAO, and Centro Internacional de Agricultura Tropical - CIAT. 2005. Gridded Population of the World, Version 3 (GPWv3): Population Count Grid. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC). http://dx.doi.org/10.7927/H4639MPP.



6

did not alter the final results of the model.5 Lastly, roads classified as “unspecified,” “local/urban,” and

“trail” were reclassified and weighted as tertiary road segments, because the speed limit of these roads

normally correlates to tertiary roads.

Fig. 2 Roadway Network

Rail Network

The rail network (fig. 3) is a composite of two data sources. There are no comprehensive datasets of

railways throughout the world or the study area. As such, START utilized both OpenStreepMap6 and

EuroGraphics7 datasets to create a model of rail travel in the study area. Due to differences in gauge,

exposure to conflict, lack of funding, age of rail lines, and consistent loss of valuable construction

materials to thieves, large swaths of the study region in northern Africa lack integrated rail

infrastructure.8

5 This model serves as a proof of concept that it is possible to create a multimodal network to investigate flows of people over vast distances. The lack topological accuracy is an issue at the local level, particularly in Europe where overpasses, underpasses, and unidirectional streets are common. 6 OpenStreetMap contributors. Openstreetmap.org 7 EuroRegionalMap v.8. EuroGraphics. http://www.eurogeographics.org/products-and-services/euroregionalmap 8 Collier, Paul, and James Cust. "Investing in Africa’s infrastructure: Financing and policy options." Annu. Rev. Resour. Econ. 7, no. 1 (2015): 473-493.



7

Fig. 3 Rail Network

Maritime Network

The ferry network (fig. 4) is a composite of two data sources. START utilized both EuroGraphics9 and

hand-digitized ferry routes derived from GoogleMaps. EuroGraphics data provides extensive routes for

ferry traffic between various European, North Africa, and Levant countries. Any routes that were missing

from this dataset were supplemented with the Google Maps digitized routes in order to create a robust,

comprehensive ferry network of the study region.

9 EuroRegionalMap v.8. EuroGraphics. http://www.eurogeographics.org/products-and-services/euroregionalmap



8

Fig. 4 Ferry Network

Illicit Maritime

The illicit maritime network (fig. 5) is comprised of hand-digitized lines from Turkey to Greek islands

known to be transited by migrants and foreign fighters entering Greece via Aegean Sea routes. Based off

of UNHCR reports, these Greek islands are not recognized as official reception centers for migrants but

have encountered a high volume of human traffic in 2015.10 Once in the European Union, illicit transit

routes (i.e., human trafficking) highly correlate to licit modes of transit.11

10 UNHCR. Greece- Sea Arrivals 2015. Jan. 1, 2016. The UN Refugee Agency. http://data.unhcr.org/mediterranean/download.php?id=442 11 Ruggiero, V. (2015). Organised crime in Europe and the mafia method. European Police Science and Research Bulletin, (13), 15-21. Harrison, David (2015). “Britain’s modern slave trade,” Al Jazeera.



9

Fig. 5 Illicit Maritime Network

Network Speeds

While obstacles to efficient transit across a transportation network can slow down travel, the

transportation modes have inherent speeds governed by the capabilities and limitations of that particular

mode. For the entire transportation network, speed in miles per hour (MPH) was calculated based on a

review of available speeds for the various modes of transit used (table 1).12 Once the speed is attributed

to each mode of transportation, time is calculated in minutes for specific segments.

Mode of Transportation Speed

(Mph)

Highway Road 60

Primary Road 50

Secondary Road 40

Tertiary Road 20

Railroad 30

Airline 450

Ferry 25

Illicit Maritime 25

12 Roads classified as Local/Urban, Trail, or Unspecified were all considered Tertiary for our purpose.



10

Intermodal Network Transitions

The NASA global roads dataset represents a highly accurate major roads dataset; however, it lacks the

spatial accuracy to enable direct connection to other transportation networks. In order to account for

this, connection lines were created using a nearest-distance function to ensure connectivity between

roadways, rail stations, airports, ferry ports, and illicit maritime exit/entry points.

One important component of transition networks relevant to a network-based transportation model is

the consideration of waiting times while transferring from one modality to another. Every transition

between networks requires at least some period of time to adjust equipment, transfer to a new vehicle, or

load/unload material. Each transition also increases the total time required to complete a route segment,

providing important delays that could shift the selection of routes from one method of transportation to

another. As not all waiting times between two modes of transit are equivalent, START constructed a table

of stopping times between one mode and another based on a review of known waiting periods for

various transition points (table 2). In essence, these periods act as a delayed barrier to a route,

encouraging the selection of other routes that may have a longer distance but a shorter transition time.

So, each connection line is attributed with a waiting time that represents the reality of the time it takes to

transfer from one specific modality to the next.

Network Transition

Connections

Waiting

Period

Roads to International Airports 150 Minutes

Roads to Ferry Ports 30 Minutes

Roads to Rail Stations 30 Minutes

Roads to Illicit Maritime 0 Minutes Table 2: Waiting periods between nodes

The network model was built with a compilation of segments from the aforementioned transportation

networks and designed using endpoint connectivity to only allow transitions at the appropriate network

nodes. Both roadways and railroad lines were planarized to ensure complete and comprehensive

connectivity, and produce a thoroughly traversable network dataset.

Transportation Network Summary

The multiple modes of transportation, each with individual transit speeds and specific transition waiting

periods, complete a substantial network model of the possible and most probable routes an individual

would take when travelling to, or from, ISIL operational areas. The total network contains 1,488,582 line

segments, 1,195,537 junction points between line segments, and millions of possible routes connecting

origin points to destination points in the study area. Given the relative size of the network, one of the key

considerations of this report is the identification of several “chokepoints” within the network dataset, for



11

which the flow of potential foreign fighter routes is constrained to a small number of transportation hubs

and corridors. As a result, investment of interdiction efforts or intelligence gathering at these locations

could provide targeted gains in the ability to interdict foreign fighters travelling to and from the ISIL

operational areas.

Areas of Operation

Modeling Group-Specific Route Preferences

With the completion of both the network of possible routes and the specific travel time costs or

incentives to transport individuals via specific routes, the third component of this project focused on

modeling the route preferences of ISIL fighters travelling throughout the study region. A number of

factors were implemented into the model to serve as weights specific to ISIL foreign fighters that modify

the route selection preferences of the algorithm. The possible effects of these factors are dichotomous –

they either incentivize or discourage foreign fighters from taking certain routes or modalities.

Areas of Operation

One of the core influences in the selection of routes by foreign fighters traveling to ISIL-controlled

territory is the operational area of ISIL itself and the operational areas of opposition forces in the region.

Contests for control over specific routes suggest that the area where a group operates is directly tied to

its use of that area's transportation infrastructure for transporting goods and/or people. A potential

foreign fighter traveling to ISIL controlled area is going to avoid areas controlled by Kurdish or Iraqi

Security Forces, and instead prefer to use areas directly controlled by ISIL to utilize existing

transportation corridors and smuggling networks. To capture this dynamic, operational areas for the

three main forces in the study region – ISIL, Kurdish, and Iraqi Security Forces – were digitized using data

collected from the Institute for the Study of War.13

ISIL Area of Operation

Individuals attempting to travel to destinations within ISIL-controlled territory will prefer to travel

through ISIL-controlled area (fig. 6), rather than areas controlled by opposition forces. This decreases

their risk of capture. Drawing from both the operational area digitization and the network data, potential

routes in ISIL controlled territory will be traversed twice as fast to simulate route preference.

13 Martin, Patrick, Emily Anagnostos, and Rachel Bessette. Iraq Control of Terrain Map: February 9, 2016. Feb. 9, 2016. Institute for the Study of War. http://iswresearch.blogspot.com/2016/02/iraq-control-of-terrain-map-february-9.html; Unknown Author. ISIS Sanctuary: January 29, 2016. Jan. 29, 2016. Institute for the Study of War. http://www.understandingwar.org/backgrounder/isis-sanctuary-january-29-2016

http://iswresearch.blogspot.com/2016/02/iraq-control-of-terrain-map-february-9.html



12

Fig. 6 ISIL Operational Areas

Opposition Force Area of Operation14

As ISIL territory is hypothesized to facilitate the transit of individuals to ISIL-controlled destinations,

areas where opposition forces operate (fig. 7, fig. 8) would most likely increase the transit risk. To

simulate this increased risk, routes traveling through opposition force controlled territory will be

traversed at half the speed of an unbiased route.

14 START did not include areas of operation for the Free Syrian Army or al-Nusra Front to simplify the model and show proof of concept.



13

Fig. 7 Kurdish Area of Operation

Fig. 8 ISF Area of Operation

Country Corruption

START hypothesizes that an ISIL foreign fighter would choose a route that has the highest combination of

efficiency and likelihood of success. Efficiency of the route – the speed at which one can transit between

the start and end point – is critical in regards to the route selection preferences, with the other critical



14

criteria affecting route selection preference being likelihood of success. Like route efficiency, likelihood of

success has an inverse relationship with likelihood of interdiction.

To accurately reflect a preference to select routes with highest efficiency as well as highest likelihood of

success, the research team incorporated the ControlofCorruption variable from the Worldwide

Governance Indicators (WGI) dataset15 into the transportation model (fig. 9). The ControlofCorruption

variable reflects perceptions of the extent to which public power is exercised for private gain, including

both petty and grand forms of corruption, as well as “capture” of the state by elites and private interests.

The variable is coded as a categorical variable ranging from low, moderately low, moderately high, high,

and neutral for 215 economies worldwide from 1996-2013. In incorporating the ControlofCorruption

variable into the routing algorithm as the proxy for likelihood of success, the research team hypothesized

that individuals avoiding apprehension or detection would prefer to select those routes that traverse

through countries with higher levels of corruption since it will lessen the likelihood of interdiction due to

the potential for bribes and other corrupt activities, thus increasing the likelihood of success.

In the model, the edges of the network transiting through a corrupt country received a decrease of time

by a factor of two. For example, an individual would be able to traverse through a country with the

ControlofCorruption value of high or moderately high two times faster than it could through a country

with the ControlofCorruption value of neutral. For countries with low or moderately low values travel time

takes two times as long to traverse the country.

Fig. 9 Levels of State Corruption

15 The World Bank Group. “Worldwide Governance Indicators.” 2014. http://info.worldbank.org/governance/wgi/index.aspx#home.

http://info.worldbank.org/governance/wgi/index.aspx#home



15

ISIL Smuggling Districts

It’s widely accepted that ISIL engages in the smuggling of both licit and illicit material across the Iraq,

Syria, and Turkish borders. To identify these areas, START geocoded the locations of news reports and

interviews regarding the smuggling of goods, materials, or people contributing to ISIL operations. After

collecting the point locations of where these reports and interviews took place, the administrative

districts that these points fall in are then selected and implemented into the model (fig. 10) as a

preference weight. The network edges that fall within these areas received a decrease of time by a factor

of two. For example, an individual traversing a network segment that falls within a smuggling district will

travel that segment twice as fast compared to a segment that does not fall within a smuggling district.

Fig. 10 ISIL Smuggling Districts

Refugee Camps

The model assumes that ISIL will use the flow of economic migrants and asylum-seekers as a way of

obscuring the movement of militants into the European Union. Blending in with the mass movement of

refugees/migrants into and throughout Europe likely decreases an individual's likelihood of being

detected.16 To account for this in the model, any network segment that falls within five miles of a known

16 Some of the assailants involved in the November 2015 terrorist attack on Paris successfully used fake Syrian passports to hide amongst the flow of refugees. http://www.wsj.com/articles/paris-stadium-attacker-entered-europe-via-greece-1447698583



16

migrant camp (fig. 11) received a decrease in time by a factor of two. Data on migrant camps was culled

from an open-source database run by a European migrant rights group.17 For example, an individual

traversing a segment that falls within a refugee buffer zone can travel twice as fast compared to a

segment that does not fall within the buffer zone.

Fig. 11 Known Refugee Camps in the Study Area

Unofficial Border Crossings

The unofficial border crossings were determined by a UNHCR report citing locations where significant

numbers of migrants are choosing to use unofficial locations to cross country borders. Although these

locations are primarily located on the Bulgarian border with Turkey, the report also suggests other

unofficial crossings throughout Greece, Syria, Iran, and Iraq. These unofficial border locations are

referred to as green borders.18 After identifying the known green border locations (fig. 12), START

generated border crossing points where roads intersected with that country’s border and deemed that

location as an unofficial border crossing. A weight is then applied to any network edges that intersect

with that administrative area to reduce the time by a factor of two. For example, an individual traversing

a network segment that falls within an unofficial border crossing district will traverse that segment twice

as fast.

17 Close the Camps. Mapping the Migrant Camps. Migreurope. Closethecamps.org. 18 UNHCR, Mixed Migration Flows in the Mediterranean and Beyond, January 2016, http://data.unhcr.org/mediterranean/download.php?id=442



17

Fig. 12 Known Unofficial Border Crossings

Illicit Greek Island Crossings

Illicit Greek island crossings were determined from a UNHCR report of landing areas on Greek isles that

are not official reception areas.19 These isles were used as a preference weight in the routing algorithm to

represent the mass flux of migrants traveling through the Aegean Sea to seek asylum in Greece (fig. 13).

As with the other weights, any network segment traveling through these islands will be traversed twice

as fast compared to a segment that does not fall within one of these isles.

19 UNHCR. Greece- Sea Arrivals 2015. Jan. 1, 2016. The UN Refugee Agency. http://data.unhcr.org/mediterranean/download.php?id=442



18

Fig. 13 Greek Islands Known for Illicit Traffic

Border Barriers

The border barriers file was digitized using data collected from The Economist.20 The available

information includes whether the status of the border barrier is proposed, under construction, or

completed (fig. 14). For the purpose of our model, all the barriers, no matter what the current status,

were implemented as restrictions in the model to represent how an individual would travel throughout

the region when avoiding all highly monitored border crossings. However, to enable an individual to still

exploit land routes on the Turkish, Iraq, and Syria borders where a border barrier is located, START

eliminated border barrier lines if the barrier was located adjacent to previously determined unofficial

border crossing or known smuggling districts. This aims to simulate how an ISIL foreign fighter would

exploit established smuggling networks and controlled territories when attempting to emigrate from the

region undetected.

20 Élisabeth Vallet, Josselyn Guillarmou, and Zoé Barry, Raoul-Dandurand Chair, University of Quebec in Montreal (2016) Boundary Walls and Fences Worldwide. The Economist.



19

Fig. 14 Known Border Control Locations

Analysis and Results

START’s analysis began with baseline assessments of the multi-modal network model. These baseline

assessments consisted of runs of the unweighted multi-modal network. Under these conditions, the

algorithm found the unweighted optimal solution. Due to the speeds of different segments (e.g., using

airlines versus trains or automobiles), it was necessary to test the model with the air network turned on

and turned off. Likewise, the baseline model tested origin points in Europe with destinations in Syria and

Iraq and origin points in Syria and Iraq with destination in Europe. The baseline results show a proof of

concept and serve as a null hypothesis for the weighted analysis. START expects that the weighted results

will differ considerably from the unweighted baseline network.

In total, START ran 14 specific scenarios (table 3 and 4) of modeling movement between Europe and

regions in Iraq and Syria.

Scenario 1 serves as a baseline, unweighted model of routes originating in 10 Iraq and Syrian cities of

interest to 100 destination cities of interest in Europe. This first model includes the air network. Scenario

2 is identical to scenario 1 except the air network has been excluded from the model. This scenario serves

as a baseline, fastest route between Syrian and European cities. Scenario 3 is a gradual progression from

the second scenario. The air network is excluded, but weights are turned on to determine how routes

shift based on the inclusion of intervening variables. Scenario 4 shows a continued shift by excluding the



20

air network, including weights, and including border checkpoints and closings that various European

countries installed as the flow of immigrants/refugees increased.

Scenario 5 is the baseline, unweighted model originating in the selected European cities to 100 random

points in ISIL-controlled territories. START used random points because specific information related to

training camps and/or safe havens within ISIL territory is unknown. Scenario 6 is identical to scenario 1

except the air network has been excluded from the model. Scenario 7 models travel from the selected

European cities to 100 random points in Iraq and Syria, introduces weights, and provides no access to the

air network. Scenario 8 is complex in that it models from European cities to random points in Iraq and

Syria with weights, but it provides limited access to the air network. Open source sources suggest that

foreign fighters have commonly made use of commercial airlines to Turkey, and then used other modes

to enter ISIL territory; scenario 8 attempts to model this.

Scenarios 9 and 10 examine a very specific micro-section of START’s multi-modal network to test for

illicit entry into Greece from Turkey. Scenario 9 is an unweighted model, and scenario 10 is weighted to

show differences when intervening variables are introduced. Scenarios 9 and 10 are specifically meant to

model the proliferation of illicit maritime crossings between Turkey and Greece through the Aegean Sea.

Scenarios 13 and 14 are also focused on a micro-section of the multi-modal network. The specific

purpose of these runs are to route, unweighted baseline (13) and weighted (14), from Turkey to official

border crossings on the border between Greece and its European neighbors. Scenarios 13 and 14 are

designed to show land-based migration that excludes illicit island hopping in the Aegean Sea.

Scenarios 11 and 12 are designed to show how routes might shift if a foreign fighter returning from ISIL

territories originate their trip in Turkey with destinations in European cities. Scenario 11 serves as an

unweighted baseline while scenario 12 includes weights.



21

Table 3: Specific route scenarios

Table 4: Start and End Points

Scenario 1:

In this unweighted baseline model from Iraq and Syria to European cities, START observes that the model

and algorithm function properly. Air routes are the least cost path. The model (fig. 15) finds the most

efficient route from the origin point to the nearest airport, flies to a European destination nearest to the

final destination and then the most efficient road route to final destination. Roadways are used only as a

means of arriving at, or departing, airports. The edge bundling visualization (fig. 16) shows that most

routes converge on London, as the airports around London are a major hub for European air traffic.



22

Fig. 15 Fig. 16

Scenario 2:

In the unweighted baseline model from Iraq and Syria to European cities, with airlines removed from the

model, START observes wide-scale usage of major roadways throughout Europe. The chokepoints with

the highest density are located in and around the Aegean (fig. 17), due to the model’s speed preference;

the least cost path is via the Aegean or Mediterranean into Europe. Once in Europe, the model quickly

switches to roadways. Likewise, from the edge bundling visualization (fig. 18), the vast majority of routes

traverse maritime routes. As this is an unweighted model, it behaves exactly as expected.

Fig. 17 Fig. 18



23

Scenario 3:

Once intervening variables are introduced, routes become less diverse, and the distribution tightens. This

scenario builds on the previous scenario by incorporating various intervening variables. The scenario has

the same number of origin and destination points, but the model’s decisions are more constrained.

Routes begin to resemble actual migrant/refugee pathways (fig. 19) with routes running through

Bulgaria and Serbia; when the present wave of migration started last summer, this was the pathway that

received the most attention until Bulgaria altered its policies and took a more hardline stance.21 Due to

constraints, the model constrains to fewer roadways in Europe compared to the unweighted model (fig.

17). The edge bundling visualization clearly shows (fig. 20) the extent of the shift once weights are

introduced. Most routes travel across the Bosporus Strait, through Greece, and then through Bulgaria and

Serbia. START observed a new preference for a land route across North Africa.

Fig. 19 Fig. 20

Scenario 4:

The introduction of physical barriers (border checkpoints, border fences, etc.) in Europe causes the

weighted baseline model to shift to the least cost path around those barriers. The border policies Croatia

and Hungary enacted had the most significant effect on route selection (fig. 21). It is here that START

observes the Balkan Route that has now become a popular pathway for migrants/refugees. The edge

bundling visualization does not show the significance of the emergence of the Balkan Route; most routes

shift to Mediterranean and/or Aegean crossings before taking pathways through the Adriatic (fig. 22).

21 Frontex, European Agency for the Management of Operational Cooperation at the External Borders of the Member States of the European Union. (2016). “Trends and Routes,” Accessed 16 January 2016 from http://frontex.europa.eu/trends-and-routes/migratory-routes-map/. Ágh, A. (2016). The Decline of Democracy in East-Central Europe: Hungary as the Worst-Case Scenario. Problems of Post-Communism, 1-11.

http://frontex.europa.eu/trends-and-routes/migratory-routes-map/

http://frontex.europa.eu/trends-and-routes/migratory-routes-map/



24

Fig. 21 Fig. 22

Scenario 5:

The unweighted baseline model without airport access predicts that foreign fighters moving into Iraq and

Syria will have a least cost preference for travel through the Adriatic and Aegean Seas (fig. 23). The edge

bundling visualization shows the narrow distribution of route choices (fig. 24) with two main routes out

of Europe.

Fig. 23 Fig. 24

Scenario 6:

As with scenario 1, when air routes are available, START observes that the most efficient pathways for

entry into ISIL territory is via commercial airlines. Airports in England serve as the hub of chokepoints

(fig. 25). The edge bundling visualization shows the most commonly used routes run between England

and points in Iraq and Syria. (fig. 26).



25

Fig. 25 Fig. 26

Scenario 7:

Once weights are introduced, routes distribute further and chokepoints change significantly (fig. 27).

However, START observes a more realistic representation of the routes recruits have taken from Western

Europe to Iraq and Syria (fig. 28). The edge bundling visualization makes clear that despite the wide

distribution of individual routes, these routes converge in Central Europe, crossing from Greece into

Turkey via the Bosporus Strait. START also observes significant land routes across North Africa that show

likely pathways into ISIL territory.

Fig. 27 Fig. 28



26

Scenario 8:

The “limited air” scenario is designed to show how recruited foreign fighters without access to airports in

Syria or Iraq, would transit toward their goal. There are some locations where the most expeditious route

remains via maritime pathways (fig. 29), however the edge bundling visualization (fig. 30) shows the

bulk of the routes converge on Istanbul and take land routes from there into ISIL territory.

Fig. 29 Fig. 30

Scenario 9:

Scenario 9 routes an unweighted baseline of illicit pathways between Turkey and Greece. This baseline

analysis of a subsection of the network specifically tests the potential of START’s illicit network

infrastructure. The results (fig. 31) shows how illicit maritime and ferry routes network through the

Aegean to form viable, probable routes for persons attempting to cross from Turkey to the European

Union. START observed the greatest density (fig. 32) of routes around the Greek islands of Lesvos and

Tinos.



27

Fig. 31 Fig. 32

Scenario 10:

The weighted, illicit network scenario (fig. 33) is visually similar to the unweighted baseline in scenario 9,

but the edge bundling (fig. 34) shows vast differences in the dispersion of routes from the previous

scenario. While the general route selection is similar we can observe that, when compared to the

unweighted model (fig. 31), the weighted model routes on the northern Aegean and southern Greek

mainland disappear while new central Aegean routes appear. In the edge bundling image (fig. 34) START

observes a higher concentration of overlapping routes illustrated in the concentration of routes

intensifying in the central region indicating a higher likelihood traffic through the central Aegean region.

Reinforcing indications of route selection through the central Aegean region is the decline in number and

intensity of chokepoints south of Athens. Instead, START notes an upswing in the number of routes going

through central Aegean island chokepoints. Tinos and Mikonos are particularly prominent as we can see

their chokepoint count ranks (app. #9 and #10) increase from 5 and 4 to 2 and 3 respectively when

comparing the unweighted to weighted models.



28

Fig. 33 Fig. 34

Scenario 11:

This unweighted baseline model examines route selection of returning foreign fighters who have

successfully entered Turkey and are attempting to gain access to the European Union. START observes

that routes diverge widely across the continent (fig. 35). The routes mainly converge in the Aegean and

Adriatic, but also along the Turkey/Greece border and in Serbia (fig. 36).

Fig. 35 Fig. 36

Scenario 12:

The weighted model (fig. 37) of routes from Turkey to European cities is very similar to the unweighted

model; however, there are significant changes to route choices due to intervening variables (fig. 38).

Weighting the model causes the algorithm to make different decisions to achieve a least cost path. While



29

START observes similar route selection, densities change. While routes through the Aegean Sea are still

preferred by the model, traversing the Adriatic is less pronounced. Land-based routes through Greece,

Serbia, and the Balkans are more prominent.

Fig. 37 Fig. 38

Scenario 13:

From the previous models START observed that there are general routes diverting through the Aegean

towards mainland Europe via Greek islands within close proximity to the Turkish coast. In order to fully

explore the tendencies of route selection, a higher level analysis focusing specifically on Aegean routes

has been conducted. However, modelling Aegean routes proves to be difficult because of the fluid nature

of maritime routes. Using the unweighted model parameters, routes are created using known unofficial

exit locations from Turkey which are connected to known entry locations on Greek islands. The starting

locations are points within Turkey, while end locations are known official exit locations in Greece and

Bulgaria that lead to the interior of Europe. In the unweighted model (fig. 39), START observed that there

are few chokepoints concentrated on mainland Greek ports and beyond. The edge bundling visualization

shows a slight preferences for maritime routes (fig. 40).



30

Fig. 39 Fig. 40

Scenario 14:

When weights are applied, START observed generally similar routes (fig. 41) to the unweighted model

above, but routes become more concentrated (fig. 42). Previously, no Greek islands had remarkable

choke points, whereas the weighted model indicates that the islands of Patmos, Lesvos, and Limnos

become waypoints in trans-Aegean routes.

Fig. 41 Fig. 42



31

Research Opportunity

The threat of foreign fighters fueling conflict is real. The historical idea of external fighters joining a battle

for ideological reasons goes back for centuries. The influence of foreign fighters fuels ideological

campaigns by inflating troop numbers and acumen in battle. While in the past, particularly in Africa in the

1960s and 1970s, “soldiers of fortune” were common, now it seems that has shifted to “soldiers of

ideology.” As new battlefronts take shape, it will become increasingly necessary to stem the tide of

trained, and untrained, foreign fighters seeking to give root to their ideological causes.

As ISIL moves into other territory, it is possible to use the TransIT tool as a way to forecast potential

realities. For this project, START’s TransIT tool was successfully adapted to show flows of people and, as

variables were introduced, it showed how routes shifted with a good degree of accuracy. START has

become adept at uncovering illicit modes of transit and modeling them to create plausible

representations of illicit travel networks. With a new target, START could forecast flows of foreign

fighters or other trafficked material before large numbers of fighters/goods arrive.

Conclusions

Overall, the model succeeds in producing variations between baseline and weighted analyses, indicating

that the model is sensitive to intervening variables (i.e., different costs, incentives, and preferences). The

ability to replicate changes in real world conditions shows that the tool has the potential to inform

decision-making if given relevant variables. The analysis generated numerous routes for general

scenarios, both baseline and weighted, and produced other scenarios which had the explicit—and

additional—purpose of accounting for changes in real world conditions (e.g., the dramatic increase in the

flow of refugees and migrants).

As U.S. government agencies attempt to prioritize their strategies to prevent trained foreign fighters from

entering the United States, the results of these analyses provide valuable insights on locations that need

close monitoring and/or engagement to prevent/reduce the potential for human trafficking in the future.

catch them coming and going: probabilistic pathway ... · in start’s analysis by running a...

Documents